Abstract
An increasing number of applications are streaming in nature. Information packets must be encoded and transmitted sequentially in real-time, and the receiver should reproduce the source stream under strict delay constraints. The study of fundamental limits and coding schemes in such communication systems is a fertile area of research. In this talk we will show how certain judiciously constructed structured codes can yield substantial gains over baseline schemes in streaming applications.
As our main result, we will present a new family of streaming-codes that achieve significant performance gains over the practically relevant Gilbert-Elliott (GE) channel model. We will discuss how a certain deterministic approximation to the GE channel provides insights into the optimality of these codes. We will also discuss the operational significance of column-distance and column-span metrics in this setup, and show that our proposed codes achieve a near optimal tradeoff between these.
Joint work with Ahmed Badr (University of Toronto), Wai-Tian Tan (Cisco Systems) and John Apostolopoulos (Cisco Systems).